On 13 December 2016 at 21:16, Andrew Jones <drjo...@redhat.com> wrote: > On Tue, Dec 13, 2016 at 10:36:01AM +0000, Peter Maydell wrote: >> This patchset adds support for the Virtualization extensions to QEMU's >> GICv3 emulation.
> I've started some testing with this. When I boot an outer (L1) kernel > with more than one cpu allocated I get > > [ 3.441908] CPU: CPUs started in inconsistent modes > [ 3.442787] ------------[ cut here ]------------ > [ 3.445434] WARNING: CPU: 0 PID: 1 at arch/arm64/kernel/smp.c:418 > smp_cpus_done+0x80/0xa0 > ... > > and KVM does not init. Booting with only one cpu I see we do start in EL2 > and KVM does init. This turns out to be a bug in QEMU's builtin PSCI implementation: the CPU_ON call is supposed to start the new CPU in the highest implemented-and-enabled NS exception level, but we were always starting it in the exception level of the calling CPU. Since Linux does its CPU_ON calls from EL1, this meant the new CPU got started at EL1 rather than EL2. I have a fix which I will put into v2 of the patchset (which I will send mostly as an encouragement to further code review since I don't think I've seen any other comments for things I need to fix...) thanks -- PMM
